Prep Time 2012

[beej67]

Either I am tripping or this is the third time this has been posted in this thread.

Its been posted three times because people keep pretending the opposite is true to try and debunk the prep time nonmyth.

 

[AE 87]

2) When you go about game by game, you are going to find deeper explanations. Can you show that our injuries didn't happen because the defense was better prepared which put us into situations more likely to cause injury? Nesbitt injury was made more likely by a defensive play (interception) that could have been because of good film preparation on part of the defense.

Furthermore, I think it's very likely the injuries (via use of depth chart stats) is included in Vegas calculation of spreads. In other words, they likely have a probabilistic model of guys further in depth chart having to contribute time and the spread depends on that. Vegas employs statisticians and that's something I would include in my model if they were paying me for their lines.

3) I only have the SEC data from the side of teams having a bye week which supports my conclusion that GT is different from the average case. If you all want to do it for ACC from the side of the team facing extra prep for 2008-2011, I would love to have a look at it.

Here is something to get you started (it needs to be first made BCS only and then extended to also include 10 and 11)
https://docs.google.com/spreadsheet/pub?key=0Ar8cqnkh36RRdHBSTWJ6RTlRVGJRTmhZRjRSSko4bWc&output=html

I don't have to show any of that. If you're going to use a game as datum to argue that extra prep time influenced the outcome, you have to provide some rationale for this besides correlation.

Each of the 3 ACC Teams who had extra prep in 2011 were in the top 5 of the ACC in rush-defense and we still outrushed their yards/rush average. The team that shut us down the most last year, UM, did not have extra prep.

 

[beej67]

I don't have to show any of that. If you're going to use a game as datum to argue that extra prep time influenced the outcome, you have to provide some rationale for this besides correlation.

Umm, ok.

"Teams try harder vs GT if they have an extra week to prepare."

Now we get to use all the games we like as datum.

 

[cyptomcat]

I don't have to show any of that. If you're going to use a game as datum to argue that extra prep time influenced the outcome, you have to provide some rationale for this besides correlation.

Each of the 3 ACC Teams who had extra prep in 2011 were in the top 5 of the ACC in rush-defense and we still outrushed their yards/rush average. The team that shut us down the most last year, UM, did not have extra prep.

The rationale has been given and beaten to death ever since the 08 Peach bowl. This is the data that actually backs up the rationale.

Your second paragraph is a good start for analysis, now do it for 4 years and report the numbers.

 

[samsgt02]

I took a preliminary look at just ACC games with teams vs. opponents who have more than a week to prepare and how they fared against the spread. The ACC average for wins against the spread was 40% with two pushes.

Coastal
VT 55%
UVA 25%
GT 44%
UM 67%
NC 29%
DU 30%

Atlantic
CU 40%
FSu 29%
NCSU 29%
WF 60%
BC 20%
MD 43%

This is using 83 ACC games from 2008 to 2011.

 

[Legal Jacket]

The numbers do not confirm that it's related to offense.

The numbers state very clearly that there's a 90% chance having an extra week helps folks vs GT than any other team, and that GT's strength of schedule over the last four years could not have been why the numbers are skewed.



Then you believe that EVERY FOOTBALL STATISTIC IS COMPLETELY WORTHLESS.

Coaches are hired and fired based on 4 years worth of win/loss records.

Therefore 4 years worth of win/loss records is the most important and significant statistic in all of football.

Pretty flawed logic. Football statistics are, in general, pretty worthless. It's why there really is no such thing as sabermetrics in football. The main problems are players health is incredibly tough to predict, playcalling is tough to predict, and most of the sample sizes are really small because there simply aren't many games. We aren't talking baseball here.

 

[cyptomcat]

good start sam, could you share list of games and the W-L records? Google docs is free.

btw, those numbers look very low. it seems like pretty good bet to bet against those teams.

 

[AE 87]

The rationale has been given and beaten to death ever since the 08 Peach bowl. This is the data that actually backs up the rationale.

Your second paragraph is a good start for analysis, now do it for 4 years and report the numbers.

Good. Since you bring up "since the 08 Peach Bowl" as you earlier brought up the AJC and ESPN articles, why don't you respond to beej and tell him that your point is prep time versus CPJ's offense rather than CPJ's presence on the sideline.

Then, explicitly post that you truly believe--based on your statistics--that we lost the UVA and VPI games last year because they had more time to prepare for CPJ's offense and not because our defense fell apart.

If you post that this is your position, I will post data to counter it. If this is not your position, then please explain the purpose of your reference to the AJC, ESPN and 08 Peach Bowl.

 

[lonestarjacket]

I don't have to show any of that. If you're going to use a game as datum to argue that extra prep time influenced the outcome, you have to provide some rationale for this besides correlation.

Each of the 3 ACC Teams who had extra prep in 2011 were in the top 5 of the ACC in rush-defense and we still outrushed their yards/rush average. The team that shut us down the most last year, UM, did not have extra prep.

Not really. If you show correlation then you can evaluate the probability that the correlation is due to something other than coincidence. You can always find a rationalization for each incidence if you look hard enough and wish real hard, but it still does not discount the correlation. (Note that the rationalization is always subjective; correlation is by its nature objective.)

Once you show correlation you only have to conclude correlation. It does not really matter whether you can prove causality. If I notice it rains every time the sky gets dark then I do not need to describe the mechanics of clouds and droplet formation to predict when to bring an umbrella.

Anecdotal evidence to the contrary does not trump statistical analysis of correlation, no matter how much you do not want the correlation to be true. You have to find a specific flaw in the analysis with counter-analysis.

There will always be counter-examples for single games to rationalize whatever you want to believe. Look at them all. Find another correlation factor that counters the supposition of prep time. (like prep time is also correlated with team strength, for example)

I think the normalization done to compare team performance against us relative to performance against all other teams is sufficiently compelling. I am convinced. Hypothesis was made, multiple analytical experiments were done, results supported the hypothesis.

You need a counter-experiment analysis to disprove that same hypothesis. But it has to be a consistent rule, not just discounting data solely because it does not suit your purpose.

 

[cyptomcat]

I have already replied to you that extra time is a factor not the only reason. You need to stop misrepresenting my position. I have not read all of beej's posts in this thread and I don't have to defend his claims.

 

[lonestarjacket]

I have already replied to you that extra time is a factor not the only reason. You need to stop misrepresenting my position. I have not read all of beej's posts in this thread and I don't have to defend his claims.

It is a fundamentally flawed counter anyway.

Take a less controversial supposition, like home field is a quantifiable advantage. Most people accept that. Sagarin's models quantify it as about three points. But you could always argue about a single game here or there where the home team actually performed worse. That is not the point. The point is that on average across all teams playing all games one can demonstrate a pattern of improved performance with that advantage. You don't even have to explain why that advantage may exist, although you could speculate about it.

Even if you speculate on cause and your speculation is shown to be false, it still does not discount the correlation.

 

[Legal Jacket]

I have already replied to you that extra time is a factor not the only reason. You need to stop misrepresenting my position. I have not read all of beej's posts in this thread and I don't have to defend his claims.

Not to mention that I think you and beej are actually taking different stances on this crap. You've definitely done enough hard work in this thread to justify not responding to other peoples' requests.

 

[AE 87]

Not really. If you show correlation then you can evaluate the probability that the correlation is due to something other than coincidence. You can always find a rationalization for each incidence if you look hard enough and wish real hard, but it still does not discount the correlation. (Note that the rationalization is always subjective; correlation is by its nature objective.)

Once you show correlation you only have to conclude correlation. It does not really matter whether you can prove causality. If I notice it rains every time the sky gets dark then I do not need to describe the mechanics of clouds and droplet formation to predict when to bring an umbrella.

Anecdotal evidence to the contrary does not trump statistical analysis of correlation, no matter how much you do not want the correlation to be true. You have to find a specific flaw in the analysis with counter-analysis.

There will always be counter-examples for single games to rationalize whatever you want to believe. Look at them all. Find another correlation factor that counters the supposition of prep time. (like prep time is also correlated with team strength, for example)

I think the normalization done to compare team performance against us relative to performance against all other teams is sufficiently compelling. I am convinced. Hypothesis was made, multiple analytical experiments were done, results supported the hypothesis.

You need a counter-experiment analysis to disprove that same hypothesis. But it has to be a consistent rule, not just discounting data solely because it does not suit your purpose.

Thanks for the long reply, but I wasn't challenging the use of statistics and how statistics can be used to move from correlation to statistically likely causal factor. I was challenging cyp's use of statitstics.

He compares the records of ACC schools against "opponents" with 7 day prep or less and against "opponents" with 8 days or more prep. That might be a valid statistical experiment if "opponent" was a static category. However, it's obviously not. Moreover, to get a higher data sample, he sums data from the last two years of Wofford and the first two years of Groh. That might be interesting if he only looked at how defenses performed against us, but he doesn't he looks at W-L -- as if our Defense and defensive coordinator don't play a role in whether or not we win or lose.

 

[AE 87]

It is a fundamentally flawed counter anyway.

Take a less controversial supposition, like home field is a quantifiable advantage. Most people accept that. Sagarin's models quantify it as about three points. But you could always argue about a single game here or there where the home team actually performed worse. That is not the point. The point is that on average across all teams playing all games one can demonstrate a pattern of improved performance with that advantage. You don't even have to explain why that advantage may exist, although you could speculate about it.

Even if you speculate on cause and your speculation is shown to be false, it still does not discount the correlation.

FWIW, the home-field advantage quantification was determined for ALL teams over a much larger data set.

This cirmcumstance is different. He posted a study that showed that extra-prep was not a significant factor generally. Yet he's arguing that it is for CPJ at GT (not for CPJ generally, so far).

There's a big difference in the size and nature of the data sets.

 

[cyptomcat]

ae, that's actually a valid criticism, but then again every year our players and coaches change a bit or a lot, so with that logic, noone should ever look at multi-year stats. If that's your position, that's ok with me.

Also I did look at how defenses performed against us, i.e. How our offense did.

 

[AE 87]

ae, that's actually a valid criticism, but then again every year our players and coaches change a bit or a lot, so with that logic, noone should ever look at multi-year stats. If that's your position, that's ok with me.

Also I did look at how defenses performed against us, i.e. How our offense did.

Don't be melodramatic. My logic does not undermine multi-year stats in general. I was very specific. I challenged using win-loss as the primary metric for judging the impact of extra time to prepare for our offense, highlighting especially that coaching changes have been known to affect that metric.

 

[cyptomcat]

He posted a study that showed that extra-prep was not a significant factor generally.

I see you say it shows something. What test did you use to determine that sample size was large enough?

Didn't this set involve stats against multiple DCs and OCs over the years? How come you accept this set then?

 

[eg1]

Ye gods, the offseason ... :rolleyes:

 

[AE 87]

I see you say it shows something. What test did you use to determine that sample size was large enough?

Didn't this set involve stats against multiple DCs and OCs over the years? How come you accept this set then?

Fine. I don't care about that study. You posted it with its claims. I repeated it.

Now, fwiw, I took your data for extra time and non-extra time GT rushing.

I then added an extra column for that team's yds/rush allowed against BCS AQ (from cfbstats). I then added a column for the Ratio of GT's actual yds/rush over Opponents yds/rush average against BCS AQ.

I then calculated the average over the 4 years.

For Teams without extra prep
Average of GTAct/OppDef Ave = 1.33
StDev = .31

For Teams with extra prep
Average of GTAct/OppDef Ave = 1.33
StDev = .31


No difference.

Hopefully this works:

https://docs.google.com/spreadsheet/ccc?key=0AgBtvZUVCL--dEFSbU9TWkwyWHBWRGxYVUZqZ3FnalE#gid=0

Oh well, I'll worry about making the spread sheet public some other time.

 

[GBurdell]

with the number of "wrinkles" in the offense, between CJM and CPJ comparing notes and film, the leg humpers won't know what hit 'um.